YouTube says that an error caused comments critical of China to auto-delete | Hacker News Reader

YouTube says that an error caused comments critical of China to auto-delete

techcrunch.com

The night before the proposal of the "National Security Law" in Hong Kong, Google censored pro-democracy searches in Hong Kong and later used the same excuse: https://news.ycombinator.com/item?id=23256470

I'd rather give Youtube benefit of doubt here, given that Youtube can't profit in China and that Google has had a long history of standing up against China. A more plausible reason for this mishap is that, as I guess, Youtube's removal system is automated and considers user complaints. The propaganda machine of Chinese government, or even China's little pinks, may have figured out a way to game the system by filing lots of complaints.

Was that confirmed to be true? I saw it pop up on Facebook, but it just said “What Google might look like in the future if we allow this to pass” (or something like that)

Yeah, unfortunately. The image I uploaded wasn't from myself, but I confirmed it first hand before posting so it made no difference which image I posted.

There’s really no excusing this, or blaming it on some obscure bug. “Yellow” in HK is very specific these days, and definitely political.

It's disturbing that YouTube has the capability of mass banning a certain point of view with just a few keystrokes, it sounds like. Wasn't built for nothing.

If there weren't spam filters, YouTube would be overrun with comments like the ones you see on local newspapers and blogs. Then nobody could use the comments section. (And despite the reputation YouTube comments have, there are some good ones. People like commenting and creators like reading them.)

YouTube IS overrun, with low-signal, copied comments for upvotes, senseless off-topic fighting, spam to go listen to someone's mixtape, etc.

It's disgusting - and beautiful. Let the votes and Report buttons decide, or just get rid of them.

The work should be in policing the Report button - not automatically deleting shit.

“A man always has two reasons for what he does—a good one, and the real one.” - JP Morgan

'What is the truth, but a lie agreed upon.' - Friedrich Nietzsche

"Claim everything. Explain nothing. Deny everything."

-Prescott Bush

Is it xenophobic to be disturbed at influence the Chinese government seems to have over international organizations and American companies? Let’s hope that doesn’t extend to American politics as well.

Narrator: it does

Edit: heck. It extends into education too. We’ve heard about a Harvard microbiologist prof who was receiving $$$ from China

I'm guessing it's because a lot of people (possibly being paid to do so) flagged many comments with these words in them, and so the ML model learned to predict that these words would lead to complaints and just deleted it right away.

My understanding is that 共匪 (communist bandit) is a historical KMT slur against the CCP and rarely used in contemporary speech. So i'm struggling to see how there would be enough organic use of the phrase to be flagged autmatically. But maybe the bots flagging it were also the bots posting it, or posting porn/covid conspiracy content with it...

I always wonder how stuff like this squares with the nearly-unanimous praise for Google's SRE book series. My colleagues in IT engineering all hold those books up as the gold standard in site reliability, but then really stupid mistakes like this surface in the news.

So, which is it? Is Google's SRE practice not all it's cracked up to be? Are they ignoring their own well-publicized advice regarding the avoidance of these blunders?

I do not want to entertain the idea that the system was working as designed and calling it a mistake is some kind of coverup, as other people in this thread have said, because the idea that a corporation the size and power of YouTube is willing to reverse unethical decisions and throw their own engineers under the bus as scapegoats is horrific.

"abuse system deletes some comments" isn't a failure mode sre concerns itself with. Buggy software can, and will get released despite whatever operational principles you have.

Sre isn't about buglessness, but reliability.

Isn't this utterly at odds with the concept of code ownership, and how SRE doesn't take systems into production without owning them first?

SRE doesn't own the code. They own the system reliability. Devs are still responsible for writing the functional code.

Certain aspects of the release process may fall under SRE's purview, but what the tool does is ultimately the responsibility of developers (and product). The SRE book covers testing in one chapter (17), and most of that chapter is about production testing.

The question you need to ask is "is the system still within in SLOs?" If so, SRE is doing their job. If you think that resulted in a bad experience, change the SLO. But I'd ask what the heck kind of SLO would have prevented this issue?

"99.9 % of spammers get to see their spam posted within 300ms"

SRE is a discipline.

Content and account moderation is a cost centre.

>“an error in our enforcement systems”

cough bs cough

"We didn't mean to turn it on YET"

What a poor excuse. Things like this doesn't happen by mistake, it requires considerable effort on the contrary.

I take it that you’ve never worked on a complex system that probably uses machine learning.

A complex blacklist of CCP-critical keywords? No.

One cpp critical keyword. You can go write a screed against the cpp in the comments if you want.

Anyway, complex systems can have unexpected behavior. Who would have thought? It’s a good thing there aren’t any examples of a some visual object recognition software classifying a dog as a cat, right?

It's multiple, actually. But that aside, if you want people to compliment on the complex machine learning beast that tech companies has built, you're missing the point. What people are upset about is the excess censorship put in place at various social media platforms. Content takedowns should always be used as a last resort.

"We apologise for the fault in the subtitles. Those responsible have been sacked."

Shields exist because flesh is weak against sharp steel, like censorship exists because authorities are weak against sharp facts.

The error here was Google burning their remaining goodwill by trying to play nice with their next biggest market, for after USA is tapped.

Hate to be that guy like "why can't we get along". I can understand when I'm the one in some camp that would be pretty "in my face"/relatable. I'm trying to see aside from IP what the US provides/would provide to China. Food maybe. Vise versa in a manner where does China need the US. If they were to somehow "strike the mainland" whether by lack of supply, etc... what would they do when US is out of the picture, use other countries? It's hard to see the "good" with regard to my initial comment I guess I am just naive... sure I have running water/access to knowledge but do other people have that, there are always bad people... it'll be interesting to see what happens. Also more on topic, seems they seal everything on their end anyway why YouTube. And with "truly" decentralized technology not biased by nationality that would be interesting. Not sure how, data pipelines, machines, etc... nothing is guaranteed to be "secure"/not tampered, encryption maybe, even that "do you personally know every single line of code" sounds conspiratorial ha.

The US has exported over a trillion dollars worth of stuff to China in the last ten years. Here is a list of the top 20 things:

    +--------------------------+-------------------+
    |       End-Use Code       |     2010-2020     |
    +--------------------------+-------------------+
    | TOTAL                    | 1,139,993,265,000 |
    | Civilian aircraft, engin | 121,975,706,000   |
    | Soybeans                 | 112,099,797,000   |
    | Passenger cars, new and  | 76,122,032,000    |
    | Semiconductors           | 59,399,078,000    |
    | Industrial machines, oth | 47,711,057,000    |
    | Plastic materials        | 37,017,622,000    |
    | Pulpwood and woodpulp    | 33,259,673,000    |
    | Medicinal equipment      | 28,111,315,000    |
    | Copper                   | 27,571,309,000    |
    | Chemicals-other          | 26,578,759,000    |
    | Measuring, testing, cont | 26,034,463,000    |
    | Chemicals-organic        | 24,673,699,000    |
    | Logs and lumber          | 21,519,618,000    |
    | Electric apparatus       | 20,903,317,000    |
    | Pharmaceutical preparati | 20,354,316,000    |
    | Other parts and accessor | 18,093,999,000    |
    | Other industrial supplie | 17,993,747,000    |
    | Aluminum and alumina     | 17,823,603,000    |
    | Industrial engines       | 17,539,761,000    |
    +--------------------------+-------------------+

So that right there, that to me seems like you want US not for just "raw materials" but their expertise in building that stuff. I'm really ignorant to the whole war/oppression thing. I get dividing people by race/controlling media and all that is bad. Are "we" just in the way from what they want. Our freedom ideals goes against their wish, etc... Go Google haha(to me).

Reply on news.ycombinator.com